[ad_1]
Offering a peek on the structure powering superior AI factories, NVIDIA Thursday launched a video that gives the primary public have a look at Eos, its newest data-center-scale supercomputer.
An especially large-scale NVIDIA DGX SuperPOD, Eos is the place NVIDIA builders create their AI breakthroughs utilizing accelerated computing infrastructure and absolutely optimized software program.
Eos is constructed with 576 NVIDIA DGX H100 methods, NVIDIA Quantum-2 InfiniBand networking and software program, offering a complete of 18.4 exaflops of FP8 AI efficiency. This technique is a sister to a separate Eos DGX SuperPODÂ with 10,752 NVIDIA H100 GPUs, used for MLPerf coaching in November.
Revealed in November on the Supercomputing 2023 commerce present, Eos — named for the Greek goddess mentioned to open the gates of daybreak every day — displays NVIDIA’s dedication to advancing AI expertise.
Eos Supercomputer Fuels Innovation
Every DGX H100 system is supplied with eight NVIDIA H100 Tensor Core GPUs. Eos incorporates a whole of 4,608 H100 GPUs.
Because of this, Eos can deal with the most important AI workloads to coach giant language fashions, recommender methods, quantum simulations and extra.
It’s a showcase of what NVIDIA’s applied sciences can do, when working at scale.
Eos is arriving on the excellent time. Persons are altering the world with generative AI, from drug discovery to chatbots to autonomous machines and past.
To attain these breakthroughs, they want greater than AI experience and growth expertise. They want an AI manufacturing unit — a purpose-built AI engine that’s at all times out there and can assist ramp their capability to construct AI fashions at scale
Eos delivers. Ranked No. 9 within the TOP500 checklist of the world’s quickest supercomputers, Eos pushes the boundaries of AI expertise and infrastructure.
It consists of NVIDIA’s superior accelerated computing and networking alongside subtle software program choices equivalent to NVIDIA Base Command and NVIDIA AI Enterprise.
Eos’s structure is optimized for AI workloads demanding ultra-low-latency and high-throughput interconnectivity throughout a big cluster of accelerated computing nodes, making it an excellent answer for enterprises seeking to scale their AI capabilities.
Primarily based on NVIDIA Quantum-2 InfiniBand with In-Community Computing expertise, its community structure helps knowledge switch speeds of as much as 400Gb/s, facilitating the fast motion of huge datasets important for coaching advanced AI fashions.
On the coronary heart of Eos lies the groundbreaking DGX SuperPOD structure powered by NVIDIA’s DGX H100 methods.
The structure is constructed to offer the AI and computing fields with tightly built-in full-stack methods able to computing at an unlimited scale.
As enterprises and builders worldwide search to harness the facility of AI, Eos stands as a pivotal useful resource, promising to speed up the journey in direction of AI-infused purposes that gasoline each group.
Editor’s word: This put up was up to date on Feb. 19, 2024, to make clear that there are two Eos methods.
[ad_2]
